Pagouro BE 1.0

Pagouro BE 1.0

A text-to-image model that draws in the Paris lithographic poster style of 1890–1910, at 512 px, on your CPU,
offline, from a USB stick. The image sibling of Pagouro, built under
the same rules: licensed and ledgered training data, numbers measured on a frozen test that ships with it,
released once, signed, timestamped.

Not from random weights. Base: CommonCanvas-S-C (Stable Diffusion 2 architecture, CC-BY-SA-4.0, trained
only on Creative Commons photographs). Our part: 3,043 public-domain posters and prints (Library of Congress,
The Met, Art Institute of Chicago), one ledger row and one labelled machine-written caption each, three
fine-tuning rounds on a rented A100 for $7.72 read from the bill. CORPUS.md has the counts.

Post-publication note (2026-10-01)

Two days after release, two independent audits of the Pagouro project (one by Grok, one by ChatGPT, both
asked "is it as FOSS as they claim?") found things we had stated too strongly or left out. We want you to
know what they found and what changed. Nothing that shipped was altered; the record was corrected around it.

• The model card did not name the model that wrote the training captions. It was Gemini 2.5 Flash-Lite
  through OpenRouter; Google's API terms restrict using outputs to develop competing models, and our reading
  is that a poster fine-tune of a Creative-Commons diffusion model is not one. The captions are the one
  training input without a public licence, and a future version can re-caption with an open-weights model.
• CommonCanvas-S-C's text encoder (OpenCLIP, trained on LAION captions) and its Stable Diffusion 2 VAE are
  inherited unchanged; they are not covered by the museum ledger or CommonCanvas's CC image set (erratum E2).
• The final selection-and-crop ledger was missing from the signed kit by a packaging slip; it is on Hugging
  Face and in the repository as ledger/training_set_r3.jsonl (erratum E1).

Follow-up (2026-10-02). Both auditors re-ran their checks after the changes and upgraded their assessments.
What they found has been fixed to the extent it can be. What remains is recorded as a limit rather than a
fix: the pretraining volume's per-shard metadata was deleted with the rented pod; the web crawl's licence
covers the dataset, not each page; and Pagouro BE's captions came from a commercial API. A future version is
the honest path for those, never an edit of 1.0.

The full record is O-49 and D-102 in docs/DECISIONS.md, and docs/ERRATA_v1.0.md.

The numbers on the box

	person (40 pictures)	judge (160 pictures)	judge agrees with the person
Subject drawn	34/40 (85 %)	131/160 (82 %)	34/40 (85 %)
Poster style	36/40 (90 %)	147/160 (92 %)	36/40 (90 %)
Asked-for words legible	0/6	2/24	6/6
Unasked lettering present	—	136/160 (85 %)	—

The untuned base draws the subject 62 % of the time under the same rubric. What it does badly: it adds
lettering to most pictures whether asked or not, and the lettering is not readable. About 95 seconds per picture on an idle 16-core desktop CPU, run from the
stick; two to three minutes when the machine is busy; about 52 minutes on an Intel N100 laptop (measured).

Download

GitHub caps a release file at 2 GiB and the model alone is 2.58 GB, so this page carries the stick in pieces.
Two ways to get the whole thing:

1. From this page: Pagouro-BE-1.0-win64-kit.zip (30 MB: runtime, documents, ledgers, evals, manifest,
   signature), plus pagouro-be-1.0-f16.gguf.part1 and .part2, plus JOIN-MODEL.bat. Unzip the kit, put
   the two parts and JOIN-MODEL.bat inside the Pagouro-BE folder, double-click JOIN-MODEL.bat: it writes
   model\pagouro-be-1.0-f16.gguf, checks its SHA-256
   (b3a394fc544ca066b83db7f4ae46e77360ccc0e3d454c1d92baa7704e4674708) and deletes the parts on a match.
2. One zip: Pagouro-BE-1.0-win64.zip — SHA-256 f6a059dd4908a29811ac16bd9dc22f89527105b0301965c58d315a9c785d15fc,
   2,609,894,682 bytes — from https://huggingface.co/Pagouro/pagouro-be-1.0 (and Arweave, if funded).

Either way VERIFY.bat must then say every listed file matches the manifest. Then, in the folder:

PAGOURO-BE "a lighthouse on a rocky coast at dusk"

The picture lands in workspace\art\. Inside: sd\ (stable-diffusion.cpp CPU build, MIT), model\pagouro-be-1.0-f16.gguf
(2.58 GB; the q8_0 quantisation of this checkpoint draws blank images on CPU, so f16 ships), ledger\,
evals\, CORPUS.md, facts.json, LICENSES.md, CHECK_YOUR_COPY.md, THREAT_MODEL.md, MANIFEST.md
(67 files) with its minisign signature and VERIFY.bat / verify_manifest.py.

Verify your copy

• Manifest SHA-256 fcbc785dfa55d84f200a532609ea23c551b654db71afe15773cc07461f2b3ec9; signed with the Pagouro
  key RWQe8tvI6RCE2uMbuILC9/rEr6bNZdcOA+WC7dHtObLE94ovGk8xuFlG (minisign -Vm MANIFEST.md -p pagouro.pub).
• OpenTimestamps proof MANIFEST_v1.0.md.ots: Bitcoin block 969239 (ots info shows the attestations; merkle roots checked against the chain).
• Weights and ledger also at https://huggingface.co/Pagouro/pagouro-be-1.0 (same hashes).
• Receipts mirrored on Quilibrium QStorage: https://pagouro.qstorage.quilibrium.com/ (be/ holds this manifest, signature, proof and facts).

Licences

Weights CC-BY-SA-4.0 (share-alike inherited from CommonCanvas, with attribution); code Apache 2.0; documents
CC BY-SA 4.0; what it draws for you: CC0, no watermark, no claim.

Eric Wade, with Claude (Anthropic). Frozen at this tag; issues are read, nothing is promised.